Sequential fear generalization and network connectivity in trauma-exposed humans with and without psychopathology

The study investigates whether trauma exposure itself, independent of formal psychiatric diagnosis, is associated with specific behavioral and neural markers of threat generalization and discrimination, and whether resilience (absence of significant psychopathology following trauma) has a distinct neural signature. Prior work has linked excessive threat generalization to disorders such as PTSD, panic disorder, and generalized anxiety disorder, but has focused largely on diagnosed PTSD and often on behavioral measures alone. Key brain regions implicated in generalization/discrimination include hippocampus, ventral and dorsomedial prefrontal cortex, insula, ACC, and thalamus, which map onto large-scale networks: default mode (DMN), salience (SN), and executive control (ECN). The present work addresses gaps by examining longitudinal changes across early and late generalization phases in trauma-exposed individuals with and without psychopathology and in non-exposed healthy controls, using network-level fMRI and behavioral risk ratings. The authors hypothesized elevated behavioral generalization in TEPG versus TEHC/HC, and limited changes in SN/ECN connectivity across stages in TEPG versus TEHC/HC.

Extant literature shows overgeneralization of conditioned threat as a candidate endophenotype across anxiety- and trauma-related disorders (e.g., PTSD, PD, GAD), with meta-analytic evidence of heightened behavioral generalization in patients relative to controls. Neural substrates implicated in fear generalization/discrimination include hippocampal subregions, ventral and dorsomedial PFC, insula, ACC, and thalamus. Large-scale networks—DMN (memory-related processes), SN (salience detection and bottom-up processes), and ECN (top-down control)—are thought to underlie these computations. Prior neuroimaging work has emphasized PTSD, limiting insights into shared trauma-related mechanisms and resilience, and has rarely examined temporal dynamics (early versus late generalization). One prior study suggested elevated early, but not late, generalization in PTSD/subthreshold PTSD relative to trauma-exposed controls, but did not relate this to network-level fMRI across sequential stages.

Design: A fear generalization/discrimination fMRI task with three phases: pre-acquisition, acquisition (80% CS+ reinforcement), and generalization split into early (EG) and late (LG) stages (33% CS+ reinforcement to prevent extinction). Stimuli included CS+ (danger), two CS- (oCS-, vCS-), and three generalization stimuli (GS1–GS3) forming a size continuum between CS+ and oCS-. Participants rated shock risk (0=no risk, 1=moderate, 2=high) following red cross probes during each stimulus/ITI. Participants: Initially N=114 (TE=79, HC=35). After exclusions for incomplete tasks/shock delivery failures and non-learning criteria, final imaging sample N=88 (TE=62; HC=26). TE was subdivided into psychopathology group (TEPG; n=31) and trauma-exposed healthy controls (TEHC; n=31). Clinical assessments included SCID-5, CAPS-5, HAM-D, LSAS, PDSS, GAD-7, and SF-36. Exclusions covered severe psychiatric conditions (e.g., psychosis, bipolar), high depression (HAM-D>25), suicidality, recent substance use disorders, psychotropic medications, MRI contraindications, and pregnancy. HC excluded for any psychiatric diagnosis or qualifying trauma exposure. Imaging: 3T GE MR750 or GE PREMIER scanners with 32-channel head coil. T1 BRAVO structural imaging; T2*-weighted EPI BOLD (TR=1.3 s, TE=28 ms, FA=60°, 27 slices, 4 mm thickness). Preprocessing in SPM12/MATLAB: realignment, slice-time correction, artifact detection (ART) with CompCor denoising, normalization to MNI with 2 mm isotropic resampling, 8 mm FWHM smoothing, nuisance regression (motion, outliers, WM/CSF), covariates for scanner/age/sex. Network analysis: Group ICA using GIFT (Infomax, MDL-estimated 30 components; ICASSO with 50 runs, components retained if Iq≥0.7 and visually non-artifactual). After excluding 6 noise components, 24 ICs remained; five ICNs selected for hypothesis testing: SN, RECN, LECN, a-DMN, p-DMN (identified via template spatial correlation). Trial-wise GLM (SPM HRF) estimated β-weights per ICN and condition; β-weights entered into group statistics. Behavioral and neural metrics: Conditioning verified via paired t-tests (CS+ vs vCS-) across phases in TE and HC. Generalization learning over time assessed as delta (EG-LG) for behavioral risk ratings via repeated-measures ANOVA with factors group and stimulus-type (vCS-, oCS-, GS1–GS3, unreinforced CS+). Steepness of generalization gradients quantified by Linear Deviation Scores (LDS = ((CS+ + CS-)/2) − mean(GS1,GS2,GS3)), with higher LDS indicating stronger generalization; delta LDS (EG-LG) tested via one-way ANOVA. Neural markers assessed using two-way ANOVAs (group × stimulus-type) on ICN β-weights’ delta (EG-LG) and one-way ANOVAs on delta LDS per ICN. Multiple comparisons correction: p<0.01 across 5 networks; Bonferroni used for post hoc tests. Scanner, age, sex included as covariates of no interest. Head motion RMS compared across groups; no significant differences.

- Conditioning: Both TE and HC learned CS contingencies (pre-acquisition CS+ vs vCS- ns; acquisition and generalization significant in both groups: HC acquisition p=0.009, t=−2.84, df=25; TE acquisition p<0.001, t=−8.01, df=60; HC generalization p=0.001, t=−3.83, df=25; TE generalization p<0.001, t=−9.95, df=60). - Trauma exposure markers (TE vs HC): - Behavioral delta (EG-LG): No significant group×stimulus effects; no significant delta LDS differences behaviorally (p>0.05). - Neural delta (EG-LG): Main effect of group for SN (F=7.01, p=0.008) and RECN (F=10.25, p=0.001), with HC showing greater reduction of activity across stages than TE (TE maintained higher SN and RECN activity). Other networks ns. - ICN-specific delta LDS: Significant group difference in SN LDS (F=7.492, p=0.04, Bonferroni), driven by greater LDS reduction in HC vs TE, indicating better discrimination learning over time in HC within SN. - Resilience markers (TEPG vs TEHC vs HC): - Behavioral delta (EG-LG): No significant group×stimulus interaction. Main effect of group (F=6.69, p=0.001), with TEPG showing consistently higher risk ratings across stages than HC (p=0.045) and TEHC (p=0.005, Bonferroni), particularly during LG. - Neural delta (EG-LG): Main effect of group for SN (F=7.13, p=0.005) and RECN (F=12.74, p=0.00002, Bonferroni). TEHC maintained higher SN and RECN activity over stages compared to TEPG (SN p=0.035; RECN p=0.0006) and HC (SN p=0.001; RECN p=0.000015), who showed greater reductions. Other networks ns. - ICN-specific delta LDS: Trending group effects in SN (F=3.70, p=0.029) and RECN (F=4.04, p=0.021). SN: HC showed greater LDS reduction than TEHC (p=0.018) and TEPG (p=0.021). RECN: TEPG showed lower LDS reduction than HC (p=0.029) and TEHC (p=0.01), suggesting poorer discrimination learning associated with RECN in TEPG. Overall, TE vs HC differences indicate a trauma exposure signature of sustained SN and RECN activity and reduced SN-based discrimination learning (LDS). Resilience is characterized by TEHC maintaining higher SN/RECN activity over time but showing RECN-associated discrimination learning (higher LDS reduction) similar to HC, whereas TEPG show higher risk ratings and lower RECN LDS reduction.

Findings address whether trauma exposure per se yields identifiable behavioral and neural markers, and whether resilient trauma-exposed individuals show distinct neural profiles. Both TE and HC learned CS contingencies. However, TE exhibited less reduction over time in SN and RECN activity and reduced improvement in SN-based discrimination (LDS), suggesting a trauma exposure phenotype marked by persistent salience and executive control engagement during generalization. When parsing TE into TEPG and TEHC, TEPG showed higher risk ratings (greater overgeneralization), particularly to GS resembling CS+, consistent with impaired discrimination seen in PTSD/anxiety literature. TEHC, despite maintaining higher SN and RECN activity across stages, showed RECN-linked improvements in discrimination learning (greater LDS reduction) comparable to HC, implying compensatory top-down control that supports resilience. Thus, intact or refined engagement of RECN may facilitate discrimination among trauma-exposed individuals who do not develop psychopathology, whereas elevated SN/RECN activity without corresponding improvement in discrimination characterizes those with psychopathology. These results suggest network-level targets (SN, ECN) for interventions across trauma-related conditions.

This study identifies both a trauma exposure signature (sustained SN and RECN activity across generalization stages with poorer SN-based discrimination learning) and a resilience signature (maintained SN/RECN activity with improved RECN-associated discrimination learning) across trauma-exposed individuals regardless of specific diagnosis. Results highlight the importance of large-scale network dynamics during sequential generalization learning and suggest intervention targets within SN and ECN to enhance discrimination and reduce overgeneralization. Potential future directions include longitudinal studies to determine causality and prognostic value of these network markers, incorporation of contextual learning to probe hippocampal contributions, broader diagnostic inclusion to test generalizability, and mechanistic/therapeutic trials (e.g., noninvasive brain stimulation such as tDCS targeting prefrontal regions) to modulate SN/ECN function and improve outcomes post-trauma.

- Lack of expected limbic (e.g., hippocampal) differences, potentially due to absence of contextual manipulation in the task; future work should incorporate context. - Data acquired on two scanners; though harmonized and controlled for, combining datasets may introduce noise. - Exclusion of participants with psychosis, schizophrenia, and bipolar disorder may limit generalizability to all trauma-exposed populations. - Cross-sectional design precludes causal inference regarding whether observed neural patterns are consequences of trauma exposure or pre-existing risk factors; longitudinal designs are needed. - Non-learners were excluded based on risk rating criteria, which may bias sample toward better learners.